Platform Team Debunks Scrum Myths After Struggling With Infrastructure Sprints

An eight-engineer infrastructure platform team spent four months applying Scrum to Kubernetes-heavy work, only to find that sprint velocity metrics created a false sense of predictability. A particularly difficult sprint — derailed by a certificate rotation, flaky cloud nodes, urgent access requests, and a production database issue — saw completed points drop from an average of ~32 to just 14. The team identified key planning failures: treating externally blocked tickets as equivalent to active work, and quietly rolling unfinished cards forward instead of recording interruptions honestly. After reforming their process to include a single Sprint Goal, reserved capacity for operational work, and transparent dependency tracking, their completion rate stabilised around 73%, down from a misleading 82%. The team concluded that Scrum can surface expectations but cannot convert unknown work into known work, and that honest forecasting is more valuable than tidy sprint numbers.